The Agent Skills Directory

[COMMAND_EXECUTION]: The conversion step executes a bash command python -m markitdown "$1" -o "$OUTPUT_DIR/output.md" where $1 and $OUTPUT_DIR are derived from user input. While double-quoted, these variables are susceptible to command substitution (e.g., $(whoami)) which allows an attacker to execute unauthorized commands.
[REMOTE_CODE_EXECUTION]: Due to the lack of input validation for file and directory paths used in shell commands, an attacker can achieve arbitrary code execution on the host environment by providing crafted arguments.
[PROMPT_INJECTION]: The skill processes external PowerPoint files to generate markdown for LLM context, which introduces a risk of indirect prompt injection. Ingestion points: The PPTX file provided as $1. Boundary markers: Absent; the converted markdown is not wrapped in delimiters or accompanied by instructions to ignore embedded commands. Capability inventory: Includes shell command execution (Bash), file system access (Read/Write), and the ability to call other skills. Sanitization: None; the content extracted from the slides is not filtered or sanitized before being added to the agent context.

convert-pptx